Visual data stories are widely used in analysis and journalism for their intuitive communication of data trends. However, existing methods often depend on templates or automated analysis, limiting personalization by neglecting user intent. To address this, we propose a user-intent-driven approach for automatic data comic generation from tabular data. The method interprets natural language queries with Text-to-SQL, extracts data facts (Text-to-Fact), and maps them to comic frames. To ensure coherence, reinforcement learning generates auxiliary frames for smooth fact sequences. A quality evaluation mechanism further optimizes results. We develop an interactive system that enables users to create personalized data comics through natural language interaction. Case studies and user evaluations demonstrate the system’s effectiveness in supporting intent-driven storytelling with greater flexibility and expressiveness.
Visual clarity in virtual reality (VR) head-mounted displays and video see-through (VST) systems is often described through hardware-oriented specifications, including panel resolution, pixels per degree (PPD), and consumer labels such as “4 K” or “8 K.” Although these descriptors are useful entry points, they provide only a partial account of what users actually perceive. Perceived detail is shaped by the combined effects of rendering configuration, runtime processing, optical transfer, VST capture and image processing, human visual limits, and task demands. This scoping-informed conceptual review reframes VR/VST display assessment as an end-to-end perception problem rather than a comparison of isolated device specifications. It draws on 83 full-text review sources and includes a separate OpenAlex metadata analysis of 8253 screened records from 2016–2025 using titles, abstracts, and related metadata. The review translates this evidence into a five-stage perception pipeline and a claim-support taxonomy that matches evidence classes and evaluation methods to the claims they can support. It argues that PPD and similar values should be treated as conditional evidence rather than universal indicators, requiring information about viewing geometry, optical transfer, runtime state, observer vision, and task demands. The paper concludes with an evaluation and reporting workflow to support clearer documentation, more comparable studies, and more human-centered communication of immersive display performance.
With the progress of 3D scanning technologies, these is a growing demand on low-cost, high-quality 3D scanning solutions for cultural heritage preservation. Specifically, accurate 3D surface detail acquisition can significantly support researchers and conservators in assessing the surface damage and restoration conditions of Chinese paintings. This, in turn, facilitates further conservation efforts as well as related tasks such as digital documentation, brushstroke analysis, and other forms of technical examination. However, existing 3D scanning solutions for cultural heritage preservation are often suffer from high cost, lack of user-friendliness, and relatively low accuracy. To address these limitations, in this paper, we propose a robust and low-cost structured light 3D scanning system specifically designed for traditional Chinese paintings. The system is built entirely from off-the-shelf consumer-grade hardware components. Based upon the gradient illumination, it enables high-quality surface scanning and produces high-resolution surface normal maps of scanned paintings. These normal maps can be further transferred into detailed 3D surface models. To enhance user friendliness, we also developed an interactive user interface that allows users to observe, analysis and annotate key features such as brushstrokes, damages and trace of previous restorations. To validate the effectiveness of our method, we conduct extensive experiments by scanning the real ancient Chinese paintings. Experimental results show our system’s potential to support a wide range of cultural heritage applications, including brushstroke analysis, damage assessment, and painting restoration.
Virtual Reality (VR) can enhance qualitative understanding of data by making measures and scales experiential, but often sacrifices quantitative precision. Conversely, conventional 2D visualizations support accurate quantitative analysis but offer limited embodied understanding. We present BridgeVis, a hybrid visualization approach that links interactive 2D charts with immersive 3D environments to support both quantitative analysis and qualitative understanding. We define a three-dimensional design space comprising interaction mode, linking mechanism, and mapping direction, and demonstrate it through four prototypes using physical and abstract datasets. An exploratory user study (N = 12) showed that adding interactive 2D charts did not significantly increase cognitive workload. Participants viewed the charts as analytical anchors for comparison, filtering, and navigation, while the 3D environments supported experiential understanding of scale, movement, and context. These findings demonstrate the potential of coordinated 2D–3D representations and inform the design of future immersive data visualization systems.
Energy efficiency diagnosis is pivotal for ensuring the economic operation of thermal power units. However, existing methodologies often struggle to effectively assimilate unstructured operation and maintenance knowledge and suffer from limited interpretability. While Large Language Models demonstrate formidable reasoning capabilities, their direct deployment in high-stakes industrial environments is impeded by severe challenges, primarily domain knowledge barriers and model hallucinations. To address these issues, we propose an intelligent energy efficiency diagnosis system that synergizes LLMs with visual analytics. First, we construct a collaborative dual- agent workflow: a Diagnostic Agent employs context-aware Retrieval-Augmented Generation to perform bottom-up attribution reasoning along the energy efficiency indicator hierarchy; meanwhile, an Evaluation Agent implements a three-dimensional scoring framework encompassing Data Grounding, Causal Consistency, and Knowledge Application to automatically validate diagnostic conclusions and mitigate hallucinations. Second, to tackle information overload during complex analysis, we design a visual analytics interface featuring Circular Glyphs to intuitively visualize the metric hierarchy and anomaly distribution. Furthermore, an “Evidence Chain” is employed to structurally manage fragmented insights, ultimately facilitating the generation of fully traceable diagnostic reports. Quantitative evaluations and expert interviews demonstrate that our system effectively integrates heterogeneous knowledge, significantly enhances diagnostic efficiency, and bolsters user trust in model outputs.
This paper addresses the challenge of multi-class unsupervised visual anomaly detection in unconstrained environments, aiming to enhance the robustness of existing frameworks against complex and uncontrolled background variations. While current anomaly detection approaches excel in controlled industrial settings, they often struggle in open and cluttered scenes, such as power-transmission facility inspection, where diverse background elements are difficult to model and are frequently misclassified as anomalies. Conventional solutions that rely on auxiliary segmentation models typically incur substantial computational overhead and introduce heavy dependencies on external systems, limiting their practicality in real-world deployments. To overcome these limitations, we propose a novel framework that leverages the intrinsic foreground-background discriminability of pre-trained DINO encoders to inject semantic priors directly into the representation learning process. Specifically, we integrate a lightweight classification head to the frozen encoder to efficiently extract foreground priors while introducing negligible inference latency. These priors are then used to guide three complementary mechanisms that explicitly decouple foreground modeling from background interference. First, we introduce a foreground-aware probabilistic masking strategy at the bottleneck, encouraging the model to allocate its reconstruction capacity to foreground semantics. Second, we impose an inter-region attention constraint within the decoder to explicitly penalize attention weights directed from foreground tokens to background distractors. Third, we propose a foreground-centric hard-patch reconstruction objective that prioritizes difficult foreground regions during optimization while dynamically suppressing background-induced anomaly scores. Extensive experiments on multiple benchmarks demonstrate that our method significantly improves detection accuracy and robustness in complex outdoor scenarios, offering a computationally efficient solution for complex real-world scenarios.
Heliostat aiming in solar power tower systems can be viewed as an inverse rendering problem: the orientations of thousands of heliostats are controlled to form a receiver radiative flux density distribution (RFDD) satisfying spatial and energetic objectives. Existing differentiable aiming methods commonly match prescribed target flux maps, which couple flux shape with absolute energy and must be redesigned when solar or weather conditions change. We propose ARCAim (accelerated ray tracing for continuous aiming), a hardware-accelerated differentiable ray tracing framework. It introduces a mixed objective that combines an energy scale invariant flux shape prior with an optical efficiency preservation term. Its Vulkan compute pipeline combines parallel RFDD accumulation with hardware bounding volume hierarchy (BVH) ray traversal, while Slang automatic differentiation and cached path replay provide gradients through the continuous ray-energy computation. On a typical synthesized 6282-heliostat field distributed with SolTrace, ARCAim improves average optical efficiency and RFDD uniformity over DiffNEG, the neural surrogate baseline, across three solar times. It optimizes online in 6.3s on the synthesized field and 11.9s on a real-world 27,135-heliostat field. It runs faster than DiffNEG while requiring no field-specific pretraining, and remains effective when cloud attenuation makes an absolute RFDD target stale.
Hierarchical tables are an important structure for organizing data with inherent hierarchical relationships. Existing studies have extensively explored methods for data fact exploration from tabular data. In particular, some studies have directly integrated visual data facts into the original table structure to support in-situ exploration, because embedding data facts within the table context can reduce cognitive load by minimizing attention shifts. However, embedding a large amount of extracted data facts into the limited space of hierarchical tables often leads to layout conflicts, hindering effective exploration. To address this issue, we propose an interactive exploration paradigm for hierarchical table data facts based on semantic zooming and develop an interactive visualization system, ZoomTable. The ZoomTable system employs semantic zooming as the interaction method, combined with a data-fact layout method and a data fact recommendation mechanism. This combination not only resolves layout conflicts, but also supports users in coherently exploring multidimensional data facts at different scales. A case study and a user experiment further validate the practicality and efficiency of ZoomTable in real-world data fact exploration scenarios.
Ultrasound data visualization is essential for diagnosis, anomaly detection, and doctor-patient communication. In recent years, real-time 3D ultrasound and rendering technologies have matured, and leading international ultrasound vendors such as GE Healthcare and Philips have incorporated realistic rendering techniques into ultrasound data visualization, achieving high image quality critical for both diagnostic and display purposes. However, due to the inherently low signal-to-noise ratio and occlusions in ultrasound data, generating highly realistic visualizations remains challenging. Artificial intelligence (AI) offers new opportunities to enhance ultrasound visualization by improving data reconstruction quality, speckle noise reduction, and physically based rendering. In this survey, we review AI-powered approaches across the ultrasound visualization pipeline, building on traditional techniques in data reconstruction, data preprocessing, and realistic rendering, highlighting how AI techniques address the unique characteristics of ultrasound data. Finally, we discuss emerging trends and future directions toward increasingly lifelike and intelligent ultrasound visualization.
Fact prediction aims to complete a knowledge graph by predicting missing facts based on existing facts. Reinforcement learning based approaches are the mainstream methods for fact prediction. However, existing reinforcement learning based approaches derive rules solely from reachability between entities, which may introduce logical errors to rules. These rules result in false positives when predicting missing facts, thus hindering the reliability of the knowledge graph. Identifying false positives is time-consuming and laborious, requiring the examination of numerous fact prediction results and analysis of their contextual information to judge whether suspected results conform to the real world. To help users complete the identification of false positives, this paper thus proposes a visual analytics approach. A reliability indicator is proposed to quantify the reliability of fact prediction results from the perspectives of rule generation and rule application, indicating those that are likely to be false positives. A visual interface is designed to present the subgraph information, matching rule information, and case triple information of fact prediction results, assisting users in making review decisions. A series of experiments are conducted to demonstrate the effectiveness of the approach, including a performance evaluation experiment, a case study, and a user study.
Open-vocabulary 3D scene segmentation serves as a fundamental capability of human perception in computer vision, as it enables systems to recognize and segment arbitrary objects in complex environments. However, existing approaches often struggle to generalize to unseen categories and lack the ability to jointly exploit geometric structure and semantic information. In this paper, we introduce OV3DSeg-VGGT, a novel framework that constructs a 3D scene segmentation model by combining distilled geometric priors from the pretrained visual transformer with semantic knowledge. Our method leverages temporally consistent 2D segmentation and cross-modal embeddings from CLIP to construct robust cross-view instance representations. By fine-tuning visual geometry transformer with a contrastive learning objective and introducing a CLIP-guided distillation projector, we align geometric features with semantic priors, enabling segmentation with strong generalization to novel categories. Extensive experiments show that OV3DSeg-VGGT outperforms existing state-of-the-art baselines and achieves generalization in open-vocabulary 3D segmentation.
In digital humanities (DH) and cultural heritage (CH), visualization-based storytelling (VBS) has become an important approach for structuring, interpreting, and communicating cultural data and research results. These domains differ markedly from other VBS application areas through their focus on historical phenomena; their reliance on heterogeneous and semantically ambiguous sources; their engagement with both fictional and factual constellations; their use of qualitative, interpretive, and critical perspectives; and their long-standing expertise in narrative practice and theory. As a result, DH and CH offer particularly rich opportunities for VBS, yet relevant work remains dispersed across humanities scholarship, DH and CH venues, visualization research, public application contexts, and tool-centered communities.Against this backdrop, we survey DH and CH work on story designs and VBS tools to identify trends and recurring patterns, promising practices, and open challenges. We contribute by (i) synthesizing storytelling design spaces into a framework tailored to VBS in DH and CH, (ii) mapping existing approaches to generate a field-level picture of practices and gaps, and (iii) highlighting future areas of concern and inquiry for VBS in relation to domain-specific epistemic questions. Overall, this survey seeks to consolidate an emerging community of practice and provide a shared reference point for future research.
Travel behavior analysis is crucial for various applications such as urban planning, transportation management, social behavior analysis, and business intelligence. Owing to the wide use of mobile phones, the massive cellular signaling data (CSD) provides us with an unprecedented opportunity to investigate people’s travel modes. However, it is challenging to explore such massive CSD due to its intrinsic sparsity, large size, and complexity. To fill this gap, we propose TMSeer, a visual analytics approach to help domain experts explore CSD and analyze city-level travel modes of people. Specifically, we first design an unsupervised method that combines rule-based heuristics (RBH) and Gaussian Mixture Model (GMM) to infer travel modes from CSD. We also take advantage of GMM’s characteristics to assess the uncertainty of inferred results. Then, we present novel visualizations to enable interactive multi-level exploration of CSD and in-depth analysis of travel modes: a map view with heatmaps to show the spatial distribution of urban traffic, a region view with hybrid radial diagrams based on clock and directional metaphors to display the traffic of different travel modes at the regional level; and a path view to visualize the detailed paths of different modes between regions. In particular, an enhanced Sankey diagram is designed to visualize the details of the movement, together with the novel band design to show the efficiency of travel modes in the path. We conducted two case studies and expert interviews with domain experts to evaluate our approach. The results demonstrate the effectiveness and usability of TMSeer in analyzing travel modes of people and urban mobility.
This work presents a design framework for AR attention-guiding cues in target localization tasks. Localization tasks involve identifying a target, moving toward it, and monitoring changes in the surrounding environment. Maintaining awareness of target-related information is challenging in real-world settings, where clutter, occlusion, and competing perceptual demands are present. AR provides a suitable environment for attention guidance because virtual cues can align with real-world targets and use sensory channels appropriate for the task and context. To understand how attention-guidance cues should express target-related information and remain noticeable in real-world localization settings, we conducted an ideation workshop with experts and novice users. Integrating perceptual and attentional perspectives, we analyzed ideas from the workshop and derived a design framework that structures attention-guidance cue design through four dimensions: What, How, Where, and When. We illustrate the framework through representative localization scenarios, including target finding, route following, and progress tracking. We conducted a two-stage evaluation to examine how well the framework supports real-world design and whether cues created with it help users perform localization tasks effectively.
The analysis and evaluation of students’ online learning behaviors are crucial tasks in online education. Analyzing various behaviors during the learning process and obtaining the corresponding evaluation results can help teachers understand students’ learning conditions, adjust teaching strategies on time, and enhance the quality of teaching. However, the existing methods for evaluating learning behaviors are often based on one specific dimension and have difficulty analyzing student-learning behavior data simultaneously across multiple online platforms, leading to incomplete and inaccurate analysis results. In this paper, we propose a visual analysis pipeline based on a comprehensive scoring model of diverse features of online learning process data that is capable of detecting and analyzing learning behavior anomalies in various learning scenarios. We also developed a visual analysis system that demonstrates the effectiveness of our pipeline in multi-platform and multi-scenario learning behavior analysis. We illustrate the effectiveness and usability of the system through two usage scenarios and in-depth user interviews.
Specifying visual encodings and interactions is exhausting but essential for authoring interactive visualizations. In this paper, we present G2, a customizable framework designed to support rapid generation of interactive visualizations with uniform specifications. G2 employs a data-driven grammar of graphics and defines a uniform set of interaction specifications. We discuss the design and implementation of G2 with rich examples, and a user interview to demonstrate its effectiveness. Since its first release in March 2016, G2 has undergone 360 iterations, received 12,000 stars, and supported over 22,600 related projects on Github.
The reputation of universities has drawn increasing attention in recent years, especially with the emergence of various rankings. However, despite advances in big data technologies that facilitate data collection and analysis, accurately defining and balancing factors related to university reputation and educational quality remains complex and tedious. Moreover, current educational assessment methods exhibit notable differences and controversies. In this paper, we present Iva, a human-in-the-loop Intelligent Visual Assessment system for higher education quality. This system utilizes large language models to analyze extensive multi-modal educational data, with visualization techniques incorporated to enable multi-scale exploration and interaction. Our extensive evaluations, including a carefully-designed user study and expert interviews, demonstrate the system’s potential value and provide insights for future improvements.
